"The underlying creature model appears to have been somewhat simplistic. Creature needs were manually coded with first order logic. Not sure how training was actually done, but it was unlikely to be too complicated."

Black and White was one of the reasons I tried to carefully write that so it was clear that not all the things I listed were "bad". What I recall is that Black and White failed its promises; the creature is nowhere near as smart as what was promised. But many people still had fun with the game that resulted.

I find the failures of AI are often quite entertaining; they have a different character of failure than other sorts. I also frequently enjoy the errors that word-based AI makes, because there's often some bizarre, twisted sense behind it that our brains can pick up on and enjoy. Markov Chain babble generators are the simplest example of that, but the serious work can generate a lot of great stuff too.